Analysis of the relationship between self-efficacy perception and academic achievement in science and technology literacy with artificial neural networks

The aim of this study is to examine the performance of the artificial neural network (ANN) method in classifying teacher candidates self-efficacy perceptions for science and technology literacy and their academic achievement according to their grade point averages. For this purpose, data were collected from 118 undergraduate 2nd, 3rd and 4th grade teacher candidates studying in Science Teaching Department in Burdur Mehmet Akif Ersoy University Faculty of Education. In the study conducted in a relationally browsing model and crosssectional design, teacher candidates were given that Self-sufficiency perception scale for science and technology literacy developed by Caymaz (2008) and a personal information form including age, gender, and grade point averages. In the analysis of the data, ANN method which is one of the artificial intelligence methods was applied and MATLAB R2015a program was used in the application of the method. In this study, the multi-layer forward feed back-propagation ANN model was preferred. Self-sufficiency perception expressions for age, gender, and science and technology literacy of the candidates were used as input data, while general grade averages were used as output data. ANN architecture has been selected in order to give the most successful results in accordance with the research problem. In order to examine the classification performance of the created ANN architecture, cross entropy (CE) values and error histogram graph were given. 35-15-9 ANN network architecture with a latent layer neuron number of 15 was found to have the lowest error value for the test data set. The ANN algorithm, which was created according to the results of the analysis, performed the classification of the candidates according to their self-sufficiency perceptions for science and technology literacy and their academic achievement according to their grade point averages with a 22.2% error rate.